An interactive behavior guidance control method and system based on execution feedback data

Abstract

The present application relates to the technical field of intelligent interaction, in particular to an interactive behavior guidance control method and system based on execution feedback data, the present application constructs a personalized baseline model containing spatial pressure distribution, fluctuation characteristics and action timing characteristics by collecting pressure distribution data of a user contact interface; on this basis, the spatial feature deviation degree and the time feature consistency of the current pressure distribution relative to the baseline model are calculated, the current state of the user is determined and the ergonomic constraint violation degree is evaluated; the feasible target point closest to the current state in the feasible region satisfying the constraint condition is searched, the corresponding guidance vector is generated, and the directional haptic guidance signal is output according to the user personalized parameters; at the same time, based on the execution feedback of the user to the guidance signal, the response characteristics of the user are modeled online and the response state of the user is inferred, and then the guidance parameters are adaptively adjusted, so as to realize accurate, comfortable and learning-capable interactive behavior guidance control.

Description

Technical Field

[0001] This invention relates to the field of intelligent interaction technology, specifically to an interactive behavior guidance and control method and system based on execution feedback data. Background Technology

[0002] With the rapid development of human-computer interaction technology, wearable interactive devices have been widely used in scenarios such as posture assistance and behavior guidance. These systems typically collect users' posture data through sensors and output guidance signals (such as vibration, sound, visual cues, etc.) based on preset rules or models to guide users to complete specific actions or correct poor postures.

[0003] Existing interactive guidance control methods mainly employ open-loop or simple closed-loop strategies. Open-loop methods output guidance signals of fixed strength based on a preset guidance program, without considering the user's actual response. While simple closed-loop methods can detect whether the user's actions have reached the target state, they only perform a binary "achieved / not achieved" judgment, lacking in-depth analysis of the user's execution feedback characteristics. A common problem with these methods is their inability to quantitatively assess the target object's responsiveness and execution level to the guidance signal, resulting in a lack of basis for setting the subsequent guidance strength.

[0004] Specifically, different users exhibit significantly different response characteristics to guidance signals of the same intensity. Some users are quick to react and have strong execution ability, and even weaker guidance signals can produce good results; while others require stronger or more frequent guidance to complete the action adjustment. Existing technologies use a uniform guidance intensity configuration, which may lead to over-intervention and resistance in the former, and insufficient guidance and correction failure in the latter. Furthermore, even for the same user, their response characteristics will change under different states (such as fatigue, distraction, etc.), and fixed guidance strategies are difficult to adapt to this dynamic nature.

[0005] Due to the lack of an adaptive adjustment mechanism based on execution feedback characteristics, existing technologies cannot achieve a precise match between guidance intensity and the user's actual responsiveness, resulting in low efficiency in human-machine collaboration and a poor interactive experience. Therefore, there is an urgent need for a technical solution that can dynamically adjust the guidance control strategy based on user execution feedback data.

[0006] To address this, an interactive behavior guidance and control method and system based on execution feedback data is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide an interactive behavior guidance control method and system based on execution feedback data. By constructing a personalized pressure baseline model, evaluating the spatial and temporal deviation characteristics of the current pressure distribution, determining the target point within a feasible domain that meets ergonomic constraints and generating directional tactile guidance signals, and dynamically modeling its response characteristics based on user execution feedback, adaptive and precise behavior guidance control is achieved.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] An interactive behavior guidance and control method based on execution feedback data includes:

[0010] Collect pressure distribution data from the user's contact interface and establish a personalized baseline model; calculate the spatial feature deviation and temporal feature consistency between the current pressure distribution and the personalized baseline model, and output the state classification label and confidence score;

[0011] Receive state classification labels and confidence scores, assess the constraint violation degree of the current pressure distribution, search for the nearest feasible target point to the user's current state within the feasible domain that satisfies the constraints, and calculate the guiding vector from the current pressure center to the feasible target point;

[0012] The system receives the guidance vector, sets the timing interval and intensity envelope according to the pre-calibrated user-personalized parameters, and generates a directional tactile guidance signal; it monitors the movement direction of the user's pressure center in real time, and adjusts the timing interval and intensity envelope when it detects that the user is moving in the wrong direction and / or has no response for a long time.

[0013] The intensity of the directional tactile guidance signal is collected as input data and the displacement of the user's pressure center is collected as output data. The second-order system parameters of the user's response characteristics are obtained by recursive least squares method, and the user's current response state is inferred.

[0014] Preferably, the process of establishing a personalized baseline model includes: extracting the average pressure and standard deviation of fluctuation of each sensing unit from the preset initial time series data, and constructing a spatial pressure distribution benchmark map and a spatial fluctuation feature map respectively; identifying the continuous high pressure area in the pressure distribution benchmark map as the support area and recording its pressure ratio relationship; extracting the trend reversal time point after low-frequency filtering of the data, and statistically analyzing the reversal interval to obtain the duration range of the user's adjustment action.

[0015] The spatial pressure distribution baseline map, spatial fluctuation characteristic map, pressure ratio relationship, and adjustment action duration range are stored as the personalized baseline model.

[0016] Preferably, the process of calculating the spatial feature deviation and temporal feature consistency includes: based on the spatial pressure distribution benchmark map and the spatial fluctuation feature map, calculating the normalized deviation between the real-time pressure of each sensing unit in the support area and the benchmark value, and statistically analyzing the proportion of units whose normalized deviation exceeds a preset threshold as the spatial feature deviation.

[0017] Within a preset sliding time window, pressure changes in each support region are assessed to obtain the pressure change trend of each region; the standard deviation of the pressure change rate is calculated to obtain the rate stability index; when the pressure change trend of all support regions is the same and the rate stability index is lower than the preset threshold, the time feature is marked as a consistent gradual change; when the pressure change trend of the support regions is different or the rate stability index is higher than the preset threshold, the time feature is marked as an inconsistent change.

[0018] Preferred, feasible target point acquisition process includes: calculating the normalized violation degree of pressure concentration, pressure gradient distribution, pressure ratio between symmetrical regions and effective contact area ratio based on preset ergonomic constraints, and obtaining the comprehensive constraint violation degree by weighted summation;

[0019] The stress data is projected into a low-dimensional feature space, and the search is iteratively performed in the opposite direction of the violation gradient until a candidate point that satisfies the constraints and / or has the minimum violation is found. If the low-dimensional Euclidean distance between the candidate point and the current point is less than the user's historical maximum adjustment range, it is confirmed as a feasible target point; otherwise, the step size is adjusted.

[0020] Preferably, the process of calculating the guidance vector includes: directly extracting the pressure center location information from the low-dimensional feature space coordinates of feasible target points as the target pressure center location; calculating the pressure weighted center coordinates of the current pressure distribution as the current pressure center location; calculating the spatial vector from the current pressure center location to the target pressure center location to obtain the guidance direction; calculating the Euclidean distance between the current pressure center location and the target pressure center location as the deviation distance; mapping the deviation distance to a guidance intensity reference value through a piecewise saturation function; determining whether the guidance direction points to the edge of the effective contact area or a recorded unsuitable area; if it points to an unsafe area, correcting the direction to the nearest safe area, and outputting a guidance vector containing the corrected direction and intensity reference value.

[0021] Preferably, the process of generating the directional tactile guidance signal includes: selecting an activation sequence along the guidance direction in the actuator array; setting an optimal time interval and intensity envelope based on the parameter with the highest user recognition accuracy; generating drive signals for each actuator according to the activation sequence, so that the intensity envelopes of adjacent actuators overlap in time according to a preset ratio; monitoring the pressure center trajectory; increasing the interval and improving the intensity contrast when the angle between the actual movement direction and the guidance direction exceeds the limit; increasing the intensity and shortening the interval when the long-term displacement is insufficient; and / or switching to a single-point continuous vibration mode.

[0022] Preferably, the process of inferring the user's current response state includes: aligning the tactile guidance signal intensity input and the pressure center displacement output in time to construct a second-order difference equation; minimizing the prediction error using the input-output data sequence, solving the equation coefficients, and extracting damping and gain parameters that characterize the system dynamics; classifying the user state into a sluggish fatigue state, a responsive awake state, or a state of misunderstanding / resistance based on the high-low combination of the damping and gain parameters relative to their respective thresholds, and generating adjustment instructions for the control parameters accordingly.

[0023] An interactive behavior guidance control system based on execution feedback data includes:

[0024] The user status recognition module is used to collect pressure distribution data of the user's contact interface, establish a personalized baseline model, calculate the spatial feature deviation and temporal feature consistency between the current pressure distribution and the personalized baseline model, and output the status classification label and confidence score.

[0025] The feasible target point calculation module is used to receive state classification labels and confidence scores, evaluate the constraint violation degree of the current pressure distribution, search for the feasible target point closest to the user's current state within the feasible domain that meets the constraints, and calculate the guiding vector from the current pressure center to the feasible target point.

[0026] The tactile sequence encoding module is used to receive the guidance vector, set the timing interval and intensity envelope according to the pre-calibrated user personalized parameters, and generate a directional tactile guidance signal; monitor the movement direction of the user's pressure center in real time, and adjust the timing interval and intensity envelope when the user is detected to be moving in the wrong direction and / or there is no response for a long time.

[0027] The parameter co-optimization module is used to collect the intensity of the directional tactile guidance signal as input data and the displacement of the user's pressure center as output data. It obtains the second-order system parameters of the user's response characteristics through the recursive least squares method and infers the user's current response state.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] 1. This invention constructs a personalized baseline model by initially collecting time-series data on user pressure distribution. This model includes a spatial pressure distribution baseline map, a spatial fluctuation feature map, the pressure ratio relationship of support areas, and the duration range of user adjustment actions. Based on this, the spatial feature deviation and temporal feature consistency of the current pressure distribution relative to the baseline model are calculated. This method effectively eliminates interference from individual differences such as user weight, posture habits, and force application methods. It allows state classification labels and confidence scores to directly reflect changes in the user's state relative to their normal state, thereby avoiding misjudgments and omissions, and improving the stability and reliability of state recognition in continuous interaction scenarios.

[0030] 2. This invention calculates the comprehensive violation degree of multiple ergonomic constraints, such as pressure concentration, pressure gradient distribution, pressure ratio of symmetrical regions, and effective contact area ratio, to search for the nearest feasible target point to the user's current state within the feasible domain, and generates a guidance vector based on this target point. This technical solution avoids large-scale, abrupt posture or behavior corrections, ensuring that the guidance process follows the principle of "adjusting to the nearest point and gradually approaching," while making safety corrections when the guidance direction points to the edge or uncomfortable area, effectively reducing user discomfort and resistance, thereby significantly improving the naturalness and sustainability of interactive guidance.

[0031] 3. This invention uses the intensity of directional tactile guidance signals as input and the displacement of the user's pressure center as output. It introduces a recursive least squares method to model the user's response characteristics online, obtaining second-order system parameters characterizing the user's dynamic behavior. Based on these parameters, it infers whether the user is currently in different response states such as fatigue, wakefulness, or misunderstanding / resistance. This mechanism enables the system not only to sense "whether there is movement" but also to understand "how to respond" and "the quality of the response," thereby allowing for targeted adjustments to control parameters such as guidance timing intervals and intensity envelopes. This significantly improves the system's adaptability and overall control effect during long-term, multi-round interactions. Attached Figure Description

[0032] Figure 1 This invention provides a schematic flowchart of an interactive behavior guidance and control method based on execution feedback data.

[0033] Figure 2 A schematic diagram of an interactive behavior guidance control system based on execution feedback data is provided for this invention.

[0034] Figure 3 This is a schematic diagram of the user's current response status inference process provided by the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0036] This invention provides an interactive behavior guidance control method based on execution feedback data, applied to wearable posture guidance devices. The device includes a pressure sensor array, a control unit, an actuator array, and a data storage unit. The pressure sensor array is arranged at the user contact interface (e.g., inside the lumbar support belt); the control unit is an embedded microcontroller responsible for data acquisition, processing, and control logic execution; the actuator array includes 16 linear resonant motors distributed in the front, back, left, and right directions of the lumbar support belt; the data storage unit stores personalized baseline models and historical data.

[0037] Example 1:

[0038] Please see Figures 1 to 2 This invention provides an interactive behavior guidance control method based on execution feedback data, applied to an interactive behavior guidance control system based on execution feedback data. The technical solution is as follows: a user state recognition module is used to collect pressure distribution data of the user's contact interface and establish a personalized baseline model; calculate the spatial feature deviation and temporal feature consistency between the current pressure distribution and the personalized baseline model, and output a state classification label and confidence score;

[0039] The feasible target point calculation module is used to receive state classification labels and confidence scores, evaluate the constraint violation degree of the current pressure distribution, search for the feasible target point closest to the user's current state within the feasible domain that meets the constraints, and calculate the guiding vector from the current pressure center to the feasible target point.

[0040] The tactile sequence encoding module is used to receive the guidance vector, set the timing interval and intensity envelope according to the pre-calibrated user personalized parameters, and generate a directional tactile guidance signal; monitor the movement direction of the user's pressure center in real time, and adjust the timing interval and intensity envelope when the user is detected to be moving in the wrong direction and / or there is no response for a long time.

[0041] The parameter co-optimization module is used to collect the intensity of the directional tactile guidance signal as input data and the displacement of the user's pressure center as output data. It obtains the second-order system parameters of the user's response characteristics through the recursive least squares method and infers the user's current response state.

[0042] Furthermore, the process of establishing a personalized baseline model includes: extracting the average pressure and standard deviation of fluctuation of each sensing unit from the preset initial time series data, and constructing a spatial pressure distribution benchmark map and a spatial fluctuation feature map respectively; identifying the continuous high pressure area in the pressure distribution benchmark map as the support area and recording its pressure ratio relationship; extracting the trend reversal time point after low-frequency filtering of the data, and statistically analyzing the reversal interval to obtain the range of user adjustment action duration.

[0043] The spatial pressure distribution baseline map, spatial fluctuation characteristic map, pressure ratio relationship, and adjustment action duration range are stored as the personalized baseline model.

[0044] Specifically, during a preset time period in the initial stage of user use, pressure timing data of each sensing unit in the pressure sensing array is continuously collected.

[0045] Extract the average pressure value of each sensing unit to construct a spatial pressure distribution baseline map;

[0046] Calculate the time fluctuation amplitude of the pressure value of each sensing unit, obtain the fluctuation standard deviation of each sensing unit, and construct a spatial fluctuation characteristic map.

[0047] Identify several consecutive sensing areas with the highest pressure values ​​in the pressure distribution baseline map, mark them as support areas, and record the pressure ratio relationship between each support area.

[0048] After performing low-frequency filtering on the pressure time series data, the time points when the pressure change trend of each support area reverses are extracted, and the time intervals between adjacent reversal points are counted to obtain the duration range of typical user adjustment actions.

[0049] The spatial pressure distribution baseline map, spatial fluctuation characteristic map, pressure ratio relationship of support area, and typical adjustment action duration range are stored as the personalized baseline model.

[0050] The personalized baseline model is essentially a set of physiological homeostatic features of the user under non-fatigue, non-forced standard postures. Specifically, the spatial pressure distribution baseline map represents the user's static anatomical support characteristics, eliminating body shape differences; the spatial fluctuation feature map represents the inherent micro-tremor noise level of muscle tension, establishing a signal-to-noise ratio benchmark; and the pressure ratio relationship represents the center of gravity distribution habits of the skeletal structure. This model maps real-time data to a relative coordinate system with the user's own characteristics as the origin, providing a normalized comparison benchmark for abnormal state determination.

[0051] When a user first wears the device, the initial calibration phase begins. For the first 3 minutes while the user maintains a normal sitting or standing posture, the control unit continuously collects pressure values ​​from each sensor unit in the pressure sensor array at a fixed sampling frequency. For each sensor unit, the control unit calculates the arithmetic mean of all sampled values ​​as the reference pressure value for that sensor unit. The reference pressure values ​​from all 64 sensor units are combined according to their position in the array to form an 8x8 spatial pressure distribution reference map.

[0052] Simultaneously, for each sensing unit, the control unit calculates the standard deviation of the pressure value. The standard deviations of all sensing units are combined by location to form a spatial fluctuation characteristic map.

[0053] The control unit analyzes the spatial pressure distribution baseline map to identify the continuous region with the highest pressure value. Specifically, starting from the sensor unit with the highest pressure, it expands outwards, grouping adjacent sensor units with pressure values ​​exceeding 70% of the maximum pressure value into the same region. Through connected component analysis, 2 to 4 support regions are identified. The sum of the pressure values ​​of the sensor units within each support region is recorded, and the pressure ratio between each region is calculated.

[0054] The control unit performs low-pass filtering on the acquired pressure time-series data to remove high-frequency fluctuations (posture sway noise) with frequencies higher than 2 Hz. Then, it extracts the pressure change trend for each support region. Specifically, it calculates the sum of pressure values ​​from all sensing units within each support region at each moment to obtain the total pressure time-series curve for that region. It detects inflection points in the total pressure curve where the pressure changes from increasing to decreasing or vice versa. The time intervals between adjacent inflection points are then calculated; these time intervals represent the duration of a single complete posture adjustment action by the user.

[0055] The system stores the spatial pressure distribution baseline map, spatial fluctuation characteristic map, the location and pressure ratio of the support area, and the duration range of typical adjustment actions in the data storage unit, forming the user's personalized baseline model. During subsequent use, the system employs a sliding window update mechanism to fine-tune and update the baseline model.

[0056] Furthermore, the process of calculating the spatial feature deviation and temporal feature consistency includes: based on the spatial pressure distribution benchmark map and the spatial fluctuation feature map, calculating the normalized deviation between the real-time pressure of each sensing unit in the support area and the benchmark value, and statistically analyzing the proportion of units whose normalized deviation exceeds a preset threshold as the spatial feature deviation.

[0057] Within a preset sliding time window, pressure changes in each support region are assessed to obtain the pressure change trend of each region; the standard deviation of the pressure change rate is calculated to obtain the rate stability index; when the pressure change trend of all support regions is the same and the rate stability index is lower than the preset threshold, the time feature is marked as a consistent gradual change; when the pressure change trend of the support regions is different or the rate stability index is higher than the preset threshold, the time feature is marked as an inconsistent change.

[0058] Specifically, the process of calculating the spatial deviation and temporal consistency between the current pressure distribution and the personalized baseline model includes:

[0059] Read the spatial pressure distribution baseline map and spatial fluctuation feature map in the personalized baseline model to obtain the real-time pressure value of each sensing unit at the current moment;

[0060] Calculate the difference between the current pressure value and the reference pressure value of each sensing unit to obtain the pressure deviation distribution;

[0061] For each sensing unit within the support area, the ratio of the absolute value of the pressure deviation to the corresponding standard deviation of the fluctuation is calculated to obtain the normalized deviation value.

[0062] The proportion of the number of sensor units whose statistically normalized deviation exceeds a preset multiple threshold to the total number of sensor units in the support area is used as the spatial feature deviation.

[0063] Within a preset sliding time window, determine whether the pressure change in each support area is increasing or decreasing, and obtain the pressure change trend of each area.

[0064] The standard deviation of the pressure change rate in each support region within the time window is calculated to obtain the rate stability index.

[0065] When the pressure change trend of all support areas is the same and the rate stability index is lower than the preset threshold, the time feature is marked as a consistent gradual change; when the pressure change trend of support areas is different or the rate stability index is higher than the preset threshold, the time feature is marked as a non-consistent change; the preset threshold is set according to three times the standard deviation of the pressure fluctuation amplitude in the user's static state, and the sliding time window is determined according to the average spectral characteristics of the human body posture adjustment action.

[0066] When the spatial feature deviation exceeds the first preset threshold and the temporal feature shows a consistent and gradual change, the output state classification label is "intervention required" and the confidence score is calculated based on the magnitude of the spatial feature deviation and the duration of the time window.

[0067] When the spatial feature deviation does not exceed the first preset threshold or the temporal feature is inconsistent, the output state classification label is normal state or actively adjusted state.

[0068] During normal operation of the equipment, the control unit continuously collects current pressure distribution data at a fixed frequency, and executes a status determination process every 0.5 seconds.

[0069] The control unit reads the personalized baseline model from the data storage unit, including the spatial pressure distribution baseline map and the spatial fluctuation characteristic map, to obtain the real-time pressure value of each sensing unit at the current moment. For each sensing unit, the difference between its current pressure value and the baseline pressure value is calculated. The pressure deviations of all 64 sensing units are combined by location to form a pressure deviation distribution. For each sensing unit within the support area (e.g., the left and right hip areas), the ratio of the absolute value of its pressure deviation to the corresponding standard deviation of fluctuation is calculated to obtain the normalized deviation value. The control unit sets a preset multiple threshold of 2, counts the number of sensing units with a normalized deviation value exceeding 2, and calculates the proportion, i.e., the spatial characteristic deviation.

[0070] The control unit sets a sliding time window of 4 seconds, retracing the pressure data from the past 4 seconds. For each support region, it calculates the total pressure at the start and end of the time window. If the total pressure at the end is greater than the total pressure at the start, the pressure change trend for that region is marked as "increasing"; otherwise, it is marked as "decreasing"; if the difference is within the fluctuation range (less than twice the standard deviation), it is marked as "stable". The control unit then checks whether the pressure change trend markings are the same for all support regions.

[0071] For each support region, the rate of change of total pressure is calculated within a 4-second time window. Specifically, the 4-second time window is divided into eight 0.5-second sub-windows, and the change in total pressure between adjacent sub-windows is calculated, yielding eight rate values. The standard deviation of these eight rate values ​​is calculated as a rate stability index. The control unit sets the rate stability threshold to 0.3 kPa per 0.5 seconds. If the rate stability index is less than 0.3 kPa per 0.5 seconds, the rate is considered stable.

[0072] When the pressure change trend is the same in all support areas and the rate stability index of all areas is below 0.3 kPa per 0.5 seconds, the control unit marks the time characteristic as "consistent gradual change". When the pressure change trend of the support areas is different or the rate stability index of any area is above 0.3 kPa per 0.5 seconds, the time characteristic is marked as "inconsistent change".

[0073] The control unit sets the first preset threshold to 25%. When the spatial feature deviation exceeds 25% and the temporal feature shows a consistent and gradual change, the output status classification label is "intervention required".

[0074] The confidence score is calculated as follows: the base score is 60 points, and the score is increased according to the deviation of the spatial feature. For every 5 percentage points exceeding the deviation, 10 points are added, with a maximum of 20 points. The score is also increased according to the duration of the time window. For every 3 seconds exceeding the duration, 10 points are added, with a maximum of 20 points.

[0075] Specifically, the control unit sets a first preset threshold. The first preset threshold is 25%, which is the lower limit of the spatial feature deviation that triggers the determination of the state requiring intervention. Its physical meaning is: when the proportion of sensor units with a normalized deviation exceeding twice the standard deviation of fluctuation within the support area reaches 25%, the system determines that the user's posture has undergone a significant pathological shift. This threshold is set based on empirical data from research on health ergonomics posture deviation recognition and can be adjusted according to individual differences during the calibration phase. When the spatial feature deviation... Exceed Furthermore, when the time characteristic shows a consistent and gradual change, the output state classification label is "state requiring intervention." Confidence score. The calculation is performed using a piecewise accumulation method, and the formula is as follows: ,in: The incremental score is based on the deviation of spatial features from the specified range. Specifically, 10 points are added for every 5 percentage points that the deviation exceeds the first preset threshold, with a maximum of 20 points. The score is an incremental score based on the duration T (in seconds) of the time window, with 10 points added for every 3 seconds of duration exceeding the limit, up to a maximum of 20 points; the confidence score S has a maximum limit of 100 points.

[0076] When the spatial feature deviation does not exceed 25% or the temporal feature is inconsistent, the control unit outputs a status classification label as normal or active adjustment. If the temporal feature is inconsistent and the rate stability index is high, it is determined to be an active adjustment state. If the spatial feature deviation is low, it is determined to be a normal state.

[0077] Furthermore, the process of obtaining feasible target points includes: calculating the normalized violation degree of pressure concentration, pressure gradient distribution, pressure ratio between symmetrical regions, and effective contact area ratio based on preset ergonomic constraints, and then weighted summing to obtain the comprehensive constraint violation degree;

[0078] The stress data is projected into a low-dimensional feature space, and the search is iteratively performed in the opposite direction of the violation gradient until a candidate point that satisfies the constraints and / or has the minimum violation is found. If the low-dimensional Euclidean distance between the candidate point and the current point is less than the user's historical maximum adjustment range, it is confirmed as a feasible target point; otherwise, the step size is adjusted.

[0079] Specifically, the process of assessing the constraint violation rate of the current pressure distribution and searching for feasible target points includes:

[0080] Read the preset ergonomic constraints, which include the upper limit of pressure concentration, the upper limit of pressure gradient between adjacent sensing units, the range of pressure ratio between left and right symmetrical regions, and the lower limit of minimum effective contact area.

[0081] Calculate the ratio of the maximum pressure value of a single sensing unit to the total pressure in the current pressure distribution, and use this ratio as the current pressure concentration.

[0082] Calculate the pressure difference between adjacent sensing units, obtain the pressure gradient distribution, and extract the maximum value of the pressure gradient;

[0083] Identify symmetrical sensing areas and calculate the ratio of the total pressure between the symmetrical areas;

[0084] Count the number of sensing units whose current pressure value exceeds the preset threshold, and calculate the percentage of effective contact area;

[0085] The difference between the current pressure concentration and the upper limit of pressure concentration, the difference between the maximum pressure gradient and the upper limit of pressure gradient, the deviation of the symmetrical pressure ratio from the preset range, and the difference between the effective contact area ratio and the lower limit of the minimum area are normalized to obtain the violation components of each constraint.

[0086] The overall constraint violation score is obtained by weighted summation of the various constraint violation components.

[0087] Specifically, this refers to the degree of comprehensive constraint violation. The calculation formula is: The normalized violation degree of each component is defined as follows: Pressure Concentration Violation Degree ,in Given the current level of pressure concentration, Upper limit of pressure concentration; pressure gradient violation degree ,in The current maximum pressure gradient between neighboring cells. Gradient upper bound; violation of symmetric pressure ratio ,in This represents the pressure ratio between the current left and right symmetrical regions. For the midpoint value of the allowed range (i.e. ), For the allowable range half-width (i.e. ); Effective contact area violation ,in This represents the current percentage of effective contact area. This represents the lower limit of the area; the value range of each component is normalized to the interval [0, 1] (values ​​exceeding 1 are truncated to 1). Weighting coefficients. =0.3、 =0.4、 =0.2、 =0.1 Based on the priority of the impact of ergonomic constraints on soft tissue injury and postural imbalance, the pressure gradient has the most significant contribution to local tissue shear injury and the highest weight, which is consistent with the principle of priority of pressure and shear force established by the design principles of ergonomic work systems in ISO 9241-5 and the clinical pressure injury risk assessment scale.

[0088] The current pressure distribution data is projected into a low-dimensional feature space consisting of the pressure center location, the principal axis direction of the pressure distribution, and the pressure dispersion by calculating the pressure weighting center, the second moment eigenvector of the pressure distribution, and the degree of pressure concentration.

[0089] Starting from the current state point, perform an iterative search in the low-dimensional feature space along the opposite direction of the constraint violation gradient, moving a preset step distance in each iteration;

[0090] After each iteration, check whether all constraints are satisfied. When all constraints are satisfied for the first time, mark the point as a candidate target point. If a point that satisfies all constraints is not found after reaching the preset maximum number of iterations, select the point with the smallest constraint violation as the candidate target point.

[0091] Calculate the Euclidean distance between the candidate target point and the current state point in the low-dimensional feature space, and determine whether the distance is less than the user's historical maximum adjustment magnitude obtained from the personalized baseline model;

[0092] If the distance condition is met, the candidate target point is determined as a feasible target point; otherwise, the search step size is shortened and the iteration is restarted, or the search direction is adjusted to point to the nearest known feasible region.

[0093] In this process, to verify whether candidate points in the low-dimensional feature space truly satisfy ergonomic constraints, a reverse transformation from the low-dimensional feature vector to the high-dimensional pressure distribution is performed. Specifically, this reverse transformation process is constructed as a constrained optimization problem: while keeping the total pressure constant and ensuring that the pressure values ​​of all sensing units are non-negative, a virtual pressure distribution is sought such that its pressure center, principal axis direction, and dispersion are consistent with the current low-dimensional feature vector, while minimizing the Euclidean distance between the virtual pressure distribution and the current actual pressure distribution. By solving this optimization problem, the corresponding virtual pressure distribution is obtained, and the specific pressure gradient and local concentration are calculated accordingly.

[0094] When the status classification label indicates a state requiring intervention and the confidence score exceeds the threshold, the control unit enters the guidance target calculation process.

[0095] The control unit reads preset ergonomic constraints from the data storage unit. The constraints include: the upper limit of pressure concentration is set to 40%, that is, the pressure of a single sensing unit should not exceed 40% of the total pressure of all sensing units; the upper limit of pressure gradient between adjacent sensing units is set to 15 kPa; the pressure ratio range of the left and right symmetrical areas is set to 0.75 to 1.25; and the lower limit of the minimum effective contact area is set to 35% of the total sensing area.

[0096] The above-mentioned constraint parameters are determined based on the following: the upper limit of pressure concentration (40%) is set according to the critical proportion of tissue ischemia caused by excessive local pressure concentration in healthy sitting posture studies; the upper limit of pressure gradient between adjacent sensing units (15 kPa) is calculated based on the spacing between sensing units (approximately 1.5 cm), corresponding to the safe upper limit of the contact surface shear gradient that normal human soft tissue can withstand; the left-right symmetrical pressure ratio range of 0.75 to 1.25 corresponds to an acceptable range of lateral offset not exceeding 15%, exceeding this range will trigger compensatory muscle load; the lower limit of minimum effective contact area (35%) is determined based on clinical observation experience of pressure concentration at bony prominences when the effective support area is insufficient. All parameters can be personalized during the initial calibration stage based on individual user body shape characteristics and subjective comfort feedback.

[0097] The control unit calculates various constraints related to the current pressure distribution. First, it iterates through all sensor units, identifies the sensor unit with the highest pressure value and its pressure value, calculates the sum of all sensor unit pressure values, and determines the pressure concentration. Then, it iterates through all adjacent sensor unit pairs, calculates the absolute value of the pressure difference between each pair, and extracts the maximum pressure difference among all adjacent sensor unit pairs. Next, it identifies symmetrical regions on the left and right sides based on the sensor array's axis of symmetry (vertical center line), calculates the sum of pressure values ​​of all sensor units on the left side and the sum of pressure values ​​of all sensor units on the right side, and determines the symmetrical pressure ratio. Finally, it counts the number of sensor units with pressure values ​​exceeding 5 kPa and calculates the percentage of effective contact area.

[0098] The control unit performs a weighted summation of each violation component, with the following weights: pressure concentration 30%, pressure gradient 40%, symmetric pressure ratio 20%, and effective contact area 10%.

[0099] The control unit projects the current pressure distribution data into a low-dimensional feature space. Specifically, it first calculates the coordinates of the pressure weighting center. The horizontal axis of the pressure center is the sum of the products of the column number of all sensing units and their pressure values, divided by the total pressure. The vertical axis is the sum of the products of the row number of all sensing units and their pressure values, divided by the total pressure.

[0100] Then, the second-order matrix of the pressure distribution is calculated as follows: For each sensing unit, calculate its lateral and longitudinal offsets relative to the pressure center. The lateral offset equals the column number of the sensing unit minus the x-coordinate of the pressure center, and the longitudinal offset equals the row number of the sensing unit minus the y-coordinate of the pressure center. Calculate the square of the lateral offset multiplied by the pressure value of the sensing unit, and sum the contributions of all sensing units to obtain the element in the first row and first column of the second-order matrix. Calculate the square of the longitudinal offset multiplied by the pressure value, and sum the results to obtain the element in the second row and second column. Calculate the lateral offset multiplied by the longitudinal offset and then multiplied by the pressure value, and sum the results to obtain the element in the first row and second column (this element equals the element in the second row and first column, indicating matrix symmetry). Perform eigenvalue decomposition on this matrix to calculate two eigenvalues ​​and their corresponding eigenvectors. The direction of the eigenvector corresponding to the larger eigenvalue is the principal axis direction of the pressure distribution.

[0101] Next, the pressure dispersion is calculated. The specific method is as follows: calculate the standard deviation of the pressure values ​​of all sensing units and divide it by the average pressure value to obtain the coefficient of variation as the pressure dispersion.

[0102] The pressure center location, principal axis direction, and pressure dispersion are combined into a low-dimensional feature vector, which is a 4-dimensional vector. The coordinates of the current state point in the low-dimensional feature space are this 4-dimensional vector.

[0103] The control unit sets the search step size to 10% of the current constraint violation. The control unit calculates the gradient of the constraint violation in the low-dimensional feature space by: relatively perturbing each component of the low-dimensional feature vector, with the perturbation amount set to 1% of the current value of the component to ensure that the relative proportion of the perturbation of each component is consistent; for each component, after increasing the perturbation amount, keeping other components unchanged, recalculating the corresponding constraint violation, and obtaining the gradient in the direction of the component by dividing the change in constraint violation by the perturbation amount.

[0104] Starting from the current state point, move the step size distance in the opposite direction of the gradient. Specifically, normalize the gradient vector (divide it by its magnitude), multiply it by the step size, and subtract the offset from the current state point coordinates to obtain the new state point coordinates.

[0105] The control unit reverses the low-dimensional feature vector of the new state point into a pressure distribution, recalculates all constraint indices, and checks whether all constraints are met. If the pressure concentration, pressure gradient, symmetric pressure ratio, and effective contact area of ​​the new state point all meet the constraints, then the point is marked as a candidate target point.

[0106] If a point that satisfies all constraints is not found after iteration, the control unit sets the maximum number of iterations to 50. If the maximum number of iterations is reached and a point is still not found, the point with the lowest constraint violation during the iteration process is selected as the candidate target point.

[0107] The control unit calculates the Euclidean distance between the candidate target point and the current state point in the low-dimensional feature space.

[0108] The control unit reads the user's historical maximum adjustment range from the personalized baseline model, which is obtained by analyzing the user's past proactive adjustment actions.

[0109] If the distance does not meet the requirements, the control unit shortens the search step size to 50% of the original step size and restarts the iterative search. Alternatively, the control unit finds the closest known feasible region to the current state point based on the historical successful target points recorded in the personalized baseline model and adjusts the search direction to point towards that region.

[0110] Furthermore, the process of calculating the guidance vector includes: directly extracting the pressure center location information from the low-dimensional feature space coordinates of feasible target points as the target pressure center location; calculating the pressure weighted center coordinates of the current pressure distribution as the current pressure center location; calculating the spatial vector from the current pressure center location to the target pressure center location to obtain the guidance direction; calculating the Euclidean distance between the current pressure center location and the target pressure center location as the deviation distance; mapping the deviation distance to a guidance intensity reference value through a piecewise saturation function; determining whether the guidance direction points to the edge of the effective contact area or a recorded unsuitable area; if it points to an unsafe area, correcting the direction to the nearest safe area, and outputting a guidance vector containing the corrected direction and intensity reference value.

[0111] Specifically, the process of calculating the guiding vector includes: directly extracting the pressure center location information from the low-dimensional feature space coordinates of the feasible target point as the target pressure center location;

[0112] Calculate the pressure weighted center coordinates of the current pressure distribution, and use them as the current pressure center location;

[0113] Calculate the spatial vector pointing from the current pressure center position to the target pressure center position to obtain the guidance direction;

[0114] Calculate the Euclidean distance between the current pressure center position and the target pressure center position, and use it as the deviation distance;

[0115] The deviation distance is mapped to a guidance intensity reference value through a piecewise saturation function. When the deviation distance is less than the first distance threshold, the guidance intensity reference value is set to an extremely low value. When the deviation distance is between the first distance threshold and the second distance threshold, the guidance intensity reference value increases linearly with the deviation distance. When the deviation distance is greater than the second distance threshold, the guidance intensity reference value remains constant.

[0116] Determine whether the location pointed to by the guidance direction is within the preset range of the edge of the current effective contact area or within the area of ​​user feedback discomfort recorded in the personalized baseline model;

[0117] When the guidance direction points to an unsafe area, while maintaining the general trend indicated by the guidance direction, adjust the guidance direction to point to the nearest safe area;

[0118] The output includes a guide vector containing guide direction and guide strength reference values.

[0119] The control unit directly extracts the pressure center location information from the low-dimensional feature space coordinates of feasible target points. The first two components of the low-dimensional feature vector are the x-coordinate and y-coordinate of the pressure center. A spatial vector pointing from the current pressure center location to the target pressure center location is calculated.

[0120] Calculate the Euclidean distance between the current pressure center position and the target pressure center position. The distance is the square root of the sum of the squares of the horizontal and vertical components.

[0121] The control unit sets a first distance threshold of 0.3 cm and a second distance threshold of 2 cm. In this example, the deviation distance is 0.56 cm, which is between the first and second thresholds. The control unit uses a piecewise saturation function to map the guidance intensity reference value. When the deviation distance is less than 0.3 cm, the guidance intensity reference value is set to 10% (extremely low value). When the deviation distance is between 0.3 and 2 cm, the guidance intensity reference value is equal to 10% plus the deviation distance minus 0.3 cm, divided by 1.7 cm, and then multiplied by 70%. When the deviation distance is greater than 2 cm, the guidance intensity reference value is kept at 80% (constant upper limit). In this example, the guidance intensity reference value is 21%.

[0122] The first distance threshold (0.3 cm) is set based on 20% of the sensor unit spacing (approximately 1.5 cm). When the displacement is below this threshold, it falls within the sensor noise level, and the corresponding guidance significance is insignificant. Therefore, the guidance intensity reference value is set to an extremely low value (10%) to avoid ineffective intervention. The second distance threshold (2 cm) is set based on the upper quartile of the user's single active adjustment amplitude statistically analyzed during the initial calibration phase. Beyond this distance, further increasing the guidance intensity significantly reduces the marginal benefit of the correction effect; therefore, it is set as the saturation upper limit. Both distance thresholds can be personalized during the calibration phase based on the user's historical adjustment amplitude data.

[0123] The control unit checks the position pointed to by the guide direction (downward to the left), extends the current pressure center position along the guide direction, and determines whether the extension line passes through the edge area of ​​the effective contact area. The edge of the effective contact area is defined as: a ring of sensing units surrounding the area formed by all sensing units with pressure values ​​exceeding 5 kPa. If the extension line passes through the edge area before reaching the target pressure center position, it is determined that the guide direction is pointing to an unsafe area.

[0124] If the guidance direction points towards an unsafe area, the control unit adjusts the guidance direction by rotating it to a safe direction that is closest to the current direction and does not cross the edge area, while maintaining the main guidance trend (downward and to the left). The control unit outputs a guidance vector containing the guidance direction and guidance intensity reference values.

[0125] Furthermore, the process of generating the directional tactile guidance signal includes: selecting an activation sequence along the guidance direction in the actuator array; setting the optimal time interval and intensity envelope based on the parameter with the highest user recognition accuracy; generating drive signals for each actuator according to the activation sequence, so that the intensity envelopes of adjacent actuators overlap in time according to a preset ratio; monitoring the pressure center trajectory; increasing the interval and improving the intensity contrast when the angle between the actual movement direction and the guidance direction exceeds the limit; increasing the intensity and shortening the interval when the long-term displacement is insufficient; and / or switching to a single-point continuous vibration mode.

[0126] Specifically, the process of generating directional tactile guidance signals includes: selecting multiple actuators arranged sequentially along the direction of the guidance vector in the actuator array of the contact interface to form an activation sequence;

[0127] Read the user's personalized parameters obtained through pre-calibration. The personalized parameters include the user's recognition accuracy data for directional tactile patterns at different time intervals. Select the time interval with the highest recognition accuracy as the optimal time interval and obtain the corresponding intensity envelope shape.

[0128] The intensity level of the tactile signal is determined based on the guide intensity reference value, and the intensity level is mapped to the peak vibration intensity of each actuator.

[0129] According to the activation sequence, the start-up time is assigned to each actuator so that the start-up time interval between adjacent actuators is equal to the optimal time interval.

[0130] A vibration intensity envelope curve is generated for each actuator, so that the peak value of the envelope curve corresponds to the peak vibration intensity of each actuator, and the shape of the envelope curve conforms to the pre-calibrated intensity envelope shape.

[0131] The intensity envelopes of adjacent actuators are controlled to overlap by a preset ratio on the time axis, so that the intensity decay phase of the preceding actuator and the intensity rise phase of the following actuator partially overlap.

[0132] The generated actuator drive signals are output to the corresponding actuators to form a timing activation mode along the guiding direction;

[0133] After the tactile guidance signal is output, the movement trajectory of the pressure center is continuously monitored, and the angle between the actual movement direction of the pressure center and the guidance direction is calculated.

[0134] When the included angle exceeds the preset angle threshold, it is determined to be moving in the wrong direction, the time interval between adjacent actuators is increased and the peak intensity contrast between each actuator is improved;

[0135] When the displacement of the pressure center is less than the minimum effective displacement threshold within the preset waiting time, it is determined to be a long-term unresponsive condition. The peak vibration intensity is gradually increased and the time interval between adjacent actuators is shortened. If there is still no response, the mode is switched to a guidance mode that increases the continuous vibration time of a single actuator.

[0136] The control unit receives a guide vector, which points downward to the left, with a guide intensity reference value of 21%. The control unit selects actuators arranged sequentially along the guide direction from the actuator array. The actuator array includes 16 motors distributed on the waist support belt in a front-back, left-right manner. Based on the guide direction, the control unit selects four actuators arranged sequentially from the center position downward to the left, labeled as actuators A, B, C, and D, forming the activation sequence.

[0137] The control unit reads the user's personalized parameters from the data storage unit. These parameters are obtained through a calibration process during the user's first use. The calibration process involves applying tactile test signals at three different time intervals in different directions, and the user pressing the corresponding button to receive feedback on the perceived direction. The user's recognition accuracy is recorded for each time interval, and the time interval with the highest recognition accuracy is selected as the user's optimal time interval. This optimal time interval is determined by the user's tactile time resolution threshold obtained through a psychophysical ladder test. Simultaneously, the optimal intensity envelope shape corresponding to this time interval is recorded.

[0138] The calibration process for the optimal time interval specifically employs a modified von Békésy step-tracking method: A standardized tactile sequence is used to test the user's orientation recognition accuracy at five time intervals: 50ms, 100ms, 150ms, 200ms, and 250ms. Each interval is tested 10 times, and the time interval at which the accuracy peaks is selected as the individual's optimal time interval. If the accuracy across multiple intervals is similar (difference ≤ 5%), the longest interval is chosen to reduce actuator power consumption. The minimum effective displacement threshold (0.1 sensor unit spacing, approximately 0.2 cm) is determined based on the sensor array spatial resolution and system noise level: this threshold corresponds to 20% of the sensor unit spacing; pressure center changes below this displacement cannot be reliably distinguished from sensor measurement noise and are not considered valid user responses. This threshold can be adjusted proportionally according to specific sensor specifications.

[0139] The control unit determines the intensity level of the tactile signal based on a 21% guide intensity reference value. The control unit sets three intensity levels: low intensity (15% to 30% of the guide intensity reference value), medium intensity (30% to 60%), and high intensity (above 60%). The control unit maps the low intensity level to 45% of the peak vibration intensity of each actuator's rated output. If the rated vibration intensity of an actuator is 200 millinewtons, the peak vibration intensity is set to 90 millinewtons.

[0140] The control unit assigns start-up times to each actuator according to the activation sequence A, B, C, and D. Actuator A starts at time T0, actuator B starts at time T0 plus 100 milliseconds, actuator C starts at time T0 plus 200 milliseconds, and actuator D starts at time T0 plus 300 milliseconds.

[0141] The control unit generates vibration intensity envelope curves for each actuator, using a trapezoidal envelope based on user-defined parameters. The intensity envelope for actuator A is as follows: from 0 to 30 millineseconds, the intensity linearly increases from 0 to 90 millinewtons; from 30 to 70 millinewtons, the intensity remains at 90 millinewtons; and from 70 to 100 millinewtons, the intensity linearly decreases from 90 millinewtons to 0. Actuators B, C, and D have the same intensity envelope shape, but their start-up times are sequentially delayed by 100 milliseconds.

[0142] The control unit sets the intensity envelope overlap ratio of adjacent actuators to 50%. Specifically, the total envelope duration of actuator A is 100 milliseconds. Actuator B starts 100 milliseconds after actuator A. Therefore, the descent phase of actuator A (70 to 100 milliseconds) and the rise phase of actuator B (100 to 130 milliseconds, i.e., 0 to 30 milliseconds after actuator B starts) overlap by 30 milliseconds on the time axis, with an overlap ratio of 30 divided by 100 equaling 30%. The control unit adjusts the envelope parameters, extending the descent time to 50 milliseconds (from 70 to 120 milliseconds), so that the overlap time is 50 milliseconds, achieving an overlap ratio of 50%.

[0143] The control unit outputs the generated actuator drive signals to the corresponding actuators through the drive circuit. The actuators start to vibrate according to the set timing and intensity envelope, forming a continuous timing activation mode along the guide direction.

[0144] After the haptic guidance signal is output, the control unit continuously monitors the pressure sensor array and calculates the movement trajectory of the pressure center in real time. The control unit calculates the current position of the pressure center and compares it with the previous position to obtain the direction of movement of the pressure center. The control unit calculates the angle between the actual movement direction of the pressure center and the guidance direction, using the dot product formula. The control unit sets a preset angle threshold of 60 degrees. In this example, the angle is less than 60 degrees (0 degrees), so it is determined that the user is moving in the correct direction and the adjustment mechanism is not triggered.

[0145] If subsequent monitoring detects an angle exceeding 60 degrees, such as the pressure center moving to the upper right (opposite to the guiding direction), the control unit determines it as moving in the wrong direction. The control unit increases the time interval between adjacent actuators, extending it from 100 milliseconds to 150 milliseconds, while simultaneously increasing the peak intensity contrast between each actuator (increasing in steps of 10%-20% of the current intensity). The peak intensities of actuators A, B, C, and D are adjusted to 90, 80, 70, and 60 millinewtons, respectively, forming a decreasing intensity gradient and enhancing the clarity of direction perception.

[0146] The control unit is set to a preset waiting time of 5 seconds, and the minimum effective displacement threshold is 0.1 sensor unit spacing (approximately 0.2 cm). If the displacement of the pressure center is consistently less than 0.2 cm within 5 seconds after the tactile signal is output, the control unit determines it as a prolonged period of no response. The control unit gradually increases the peak vibration intensity from 90 millinewtons to 110 millinewtons (a 20% increase) and shortens the time interval between adjacent actuators (intensity increase step size of 15%-25%, interval shortening step size of 10ms-30ms), reducing it from 100 milliseconds to 80 milliseconds, increasing the strength and urgency of the guidance signal. If the lack of response continues for more than 10 seconds, the control unit switches to a single actuator continuous vibration mode, selecting one actuator in the guidance direction to vibrate continuously for 2 seconds, sacrificing directional precision for greater sensory certainty.

[0147] Furthermore, the process of inferring the user's current response state refers to... Figure 3 This includes: aligning the tactile guidance signal intensity input and the pressure center displacement output in time to construct a second-order difference equation; minimizing the prediction error using the input-output data sequence, solving the equation coefficients, and extracting damping and gain parameters that characterize the system dynamics; classifying the user state into a sluggish fatigue state, a responsive awake state, or a state of misunderstanding / resistance based on the high-low combination of the damping and gain parameters relative to their respective thresholds, and generating adjustment instructions for the control parameters accordingly.

[0148] Specifically, the process of inferring the user's current response state includes: recording the peak intensity of the output tactile guidance signal as the system input data sequence during each guidance cycle, and recording the displacement of the user's pressure center at the corresponding moment as the system output data sequence.

[0149] Time alignment is performed on the system input data sequence and the system output data sequence to ensure the time synchronization of input and output data;

[0150] A second-order difference equation is established to characterize the relationship between input and output. The second-order difference equation describes the linear combination relationship between the current output and the outputs of the previous two time steps, as well as the current input and the input of the previous time step.

[0151] Using the most recently preset number of input-output data pairs, the coefficients of the second-order difference equation are solved by minimizing the sum of squared errors between the model's predicted output and the actual measured output.

[0152] Specifically, the second-order difference equation takes the following form: ,in For the first Displacement of the center of pressure at any given time (unit: cm). For the first Peak intensity of tactile guidance signal at any given time (unit: millinewtons). and To output feedback coefficients, and The input gain coefficients, along with three other coefficients, collectively determine the system's dynamic response characteristics. Each coefficient is solved by minimizing the sum of squared prediction errors of the N most recent (N=20) input-output data pairs, i.e.: ;

[0153] in The model predicts the values; the matrix least squares method is used to solve for the data matrix. (No. Behavior and output vector (No. Behavior Then the coefficient vector The recursive implementation uses the recursive least squares method, which eliminates the need to recalculate the inverse each time, resulting in higher computational efficiency and meeting real-time control requirements.

[0154] The coefficients of the difference equation obtained from the solution are transformed by the discrete system transfer function to extract parameters that characterize the dynamic characteristics of the system. These parameters include frequency parameters that characterize the speed of response, damping parameters that characterize the smoothness of response, and gain parameters that characterize the sensitivity.

[0155] The user's state is classified based on the values ​​of the damping and gain parameters: when the damping parameter is higher than its corresponding first threshold and the gain parameter is higher than its corresponding second threshold, it is inferred that the user is currently in a normal state, and the guidance intensity is gradually reduced to test the user's autonomy; when the damping parameter is higher than its corresponding first threshold and the gain parameter is lower than its corresponding second threshold, it is inferred that the user is currently in a sluggish and fatigued state; when the damping parameter is lower than the first threshold and the gain parameter is higher than the second threshold, it is inferred that the user is currently in a responsive and alert state; when the damping parameter is lower than the first threshold and the gain parameter is also lower than the second threshold, it is inferred that the user is currently in a state of not understanding the guidance intention or actively resisting.

[0156] Based on the inferred user response status, adjustment instructions are generated for the confidence score trigger threshold, the step distance of the iterative search, and the peak vibration intensity of each actuator. These adjustment instructions are then transmitted to the corresponding modules through a parameter smoothing transition mechanism.

[0157] The state of not understanding the guidance intention or actively resisting is manifested in the physical signal as follows: Although the system has output a tactile guidance signal that reaches the preset intensity limit and the guidance duration exceeds the preset cognitive response window, the user's pressure center displacement is still close to zero, or the angle between the pressure center movement direction and the guidance vector direction is continuously greater than 90 degrees (i.e., moving in the opposite direction). Such abnormal input-output relationship is identified as a high impedance or negative gain mode and classified as a resistance state.

[0158] During each guidance cycle, the control unit records the peak intensity of the output tactile guidance signal as system input data, and simultaneously records the displacement of the user's pressure center at the corresponding moment as system output data.

[0159] The control unit performs time alignment on the system input data sequence and the system output data sequence. Because there is a delay in the tactile signal's journey from output to user perception and then to the generation of a motion response, the control unit compensates for this delay by analyzing the time correlation between the input signal and the output response in historical data to determine the average delay time.

[0160] The control unit establishes a second-order difference equation representing the input-output relationship. The equation is in the form that the output at the current time is equal to the first coefficient multiplied by the output at the previous time, the second coefficient multiplied by the output at the previous two times, the third coefficient multiplied by the input at the current time, and the fourth coefficient multiplied by the input at the previous time.

[0161] The control unit is preset to use the 20 most recent data pairs. Using these 20 sets of input-output data, the sum of squared errors between the predicted and actual outputs is minimized. Specifically, for each data pair, the predicted output is calculated using the difference equation, and the difference between the predicted and actual outputs is calculated. The sum of the squared differences across all 20 data pairs yields the total error. Adjusting the coefficients to minimize the total error can be achieved through matrix operations or iterative optimization algorithms.

[0162] The control unit extracts the system dynamic characteristic parameters from the coefficients of the difference equation. First, it converts the difference equation into the form of a discrete system transfer function, solves for the roots of the denominator polynomial of the transfer function, and obtains the system poles. The magnitude and argument of the poles reflect the system's frequency parameters and damping parameters.

[0163] The control unit classifies the state based on the values ​​of the damping and gain parameters, setting a first threshold (damping parameter) of 0.7 and a second threshold (gain parameter) of 0.8. In this example, the damping parameter of 0.9 is higher than 0.7, and the gain parameter of 0.56 is lower than 0.8, satisfying the condition of "damping parameter higher than the first threshold and gain parameter lower than the second threshold". The control unit infers that the user is currently in a state of fatigue with sluggish response.

[0164] If the damping parameter is below 0.7 and the gain parameter is above 0.8, for example, a damping parameter of 0.5 and a gain parameter of 1.2, it is inferred that the user is in a responsive and alert state. If the damping parameter is below 0.7 but the gain parameter is also below 0.8, for example, a damping parameter of 0.6 and a gain parameter of 0.4, it is inferred that the user is in a state of not understanding the guidance intention or actively resisting.

[0165] The control unit generates parameter adjustment instructions based on the inferred user response state. Regarding the confidence trigger threshold, the first preset threshold is reduced from 25% to 20%, making it easier for the system to trigger intervention to address the user's sluggish response. Regarding the search step distance, the step size is shortened from 10% of the constraint violation to 5%, bringing the searched target point closer to the user's current state and reducing task difficulty. Regarding peak vibration intensity, the benchmark peak intensity is increased from 90 millinewtons to 120 millinewtons, while the optimal time interval is extended from 100 milliseconds to 130 milliseconds, giving the user more time to perceive and react.

[0166] The control unit transmits adjustment commands to the corresponding modules through a parameter smoothing transition mechanism. This mechanism employs a first-order low-pass filter: the new parameter value equals the current parameter value multiplied by 0.9 plus the target parameter value multiplied by 0.1. This is updated once per control cycle, and after several cycles, the parameters smoothly transition to the target value, preventing drastic changes in system behavior caused by sudden parameter mutations.

[0167] Through the above steps, the present invention achieves complete closed-loop control from user state recognition, target point calculation, haptic guidance generation to response evaluation and parameter optimization. It can adaptively adjust the guidance strategy according to the user's real-time state, thereby improving the effectiveness of behavioral guidance and user acceptance.

[0168] Example 2:

[0169] This embodiment, based on Embodiment 1, further illustrates the specific implementation methods of the dynamic update mechanism of the personalized baseline model, the state assessment mechanism of multi-physiological parameter fusion, and the hierarchical progressive guidance strategy. This embodiment still uses the wearable posture guidance device described in Embodiment 1. In addition to including a pressure sensor array, a control unit, an actuator array, and a data storage unit, this device also integrates a miniature physiological sensor module at a specific location on the pressure sensor array.

[0170] The personalized baseline model employs a dynamic update mechanism, including:

[0171] Set the validity period of the baseline model and the confidence decay function. When the model usage time exceeds a preset proportion of the validity period, an update evaluation will be initiated.

[0172] During the update assessment period, new pressure distribution data are continuously collected, and the new data are compared with the current baseline model to calculate the systematic offset of the new data relative to the current baseline.

[0173] When the systematic offset exceeds the preset threshold for a preset number of consecutive times, it is determined that the user's habits have changed significantly, triggering the baseline model update process.

[0174] The baseline model is updated using a weighted fusion method. The historical baseline model data and the newly collected data are assigned decay weights and growth weights, respectively. The decay weights decrease as the time distance of the historical data increases, and the growth weights increase as the stability score of the new data improves.

[0175] When updating the spatial pressure distribution baseline map, a weighted average is used for the baseline pressure values ​​of each sensing unit, with the weights determined by both data timeliness and data quality.

[0176] When updating the pressure ratio relationship of the support region, the fusion update is only performed if the position of the support region in the new data is consistent with the historical model. If the position of the support region has shifted, the support region is re-labeled.

[0177] After the baseline model is updated, the parameters of the historical version of the baseline model are saved as a backup. When the updated model is detected to cause a decrease in system performance, the system is rolled back to the historical version and the update strategy parameters are adjusted.

[0178] Specifically, after establishing a personalized baseline model, a dynamic update mechanism is initiated during long-term use. When a user continuously uses the device for more than 7 days, the control unit sets the baseline model's validity period to 14 days and initiates a confidence decay assessment. The confidence decay function uses a linear function: when the model has been used for 7 days, the confidence level decreases linearly from 100%, triggering an update assessment. The update assessment period is set to 3 consecutive days. During this period, the control unit continuously collects pressure distribution data that is determined to be in a normal state. The control unit calculates the systematic offset between the new data and the current baseline model. The calculation of the systematic offset follows the following general logic: calculate the absolute value of the difference between the baseline pressure value of each sensor unit and the corresponding value in the historical model during the current assessment period, take the arithmetic mean of these differences as the average offset, and then divide the average offset by the mean of the baseline pressure of all sensor units in the historical model to obtain the relative offset percentage. The control unit sets the systematic offset threshold to 8%. When the relative offset percentage of new data exceeds 8% for 5 consecutive hours, it is determined that the user's habits have changed significantly, triggering a baseline model update. The baseline model is updated using a weighted fusion method. The growth weight of the new data is determined based on the stability score: if the spatial fluctuation standard deviation of the new data is less than 1.2 times that of the historical baseline, the stability score is high and the weight is set to 0.52; otherwise, the weight is set to 0.3.

[0179] For the spatial pressure distribution baseline map, the control unit calculates a weighted average of the baseline pressure values ​​for each sensor unit. This process is repeated for all 64 sensor units to complete the baseline map update. For the pressure ratio relationship of the support areas, the control unit first checks whether the support area positions in the new data are consistent with the historical model. The support areas are identified in the new data, the coordinates of the center positions of each area are extracted, and the offset distance from the corresponding center positions in the historical model is calculated. The control unit sets the position offset threshold to one sensor unit spacing. This threshold is based on the following considerations: minor adjustments to the user's posture during normal use will not cause the center position of the support area to shift by more than one sensor unit spacing; an offset exceeding this threshold usually indicates that the user has changed their basic sitting posture. For applications with different sensor spacings, this threshold can be adjusted proportionally; for example, the threshold is set to 3 cm when the sensor spacing is 3 cm. If the offset distance of all support areas is less than the threshold, the positions are considered consistent, and a fusion update is performed. If the support area positions have shifted, the system does not perform a fusion update but instead re-executes the area marking process, re-identifying the support areas and pressure ratio relationships based on the new data. After the baseline model update is completed, the control unit saves all historical version parameters completely in the backup area, retaining a maximum of the three most recent versions. For 24 consecutive hours after the update, the system enters performance monitoring mode, recording the false positive rate and false negative rate. When the system outputs a state requiring intervention and initiates guidance, if the user does not respond within 5 seconds and maintains the current posture for more than 1 minute, it is considered a false positive. When the system determines that the state is normal, if the user actively makes a significant adjustment within 30 seconds, it is considered a false negative. When the false positive rate exceeds 15% or the false negative rate exceeds 10% within 24 hours, performance is considered to have degraded. The system immediately rolls back to the most recent historical version and adjusts the update strategy: the weight of new data is reduced to 0.3, and the weight of historical data is increased to 0.7, making the update more conservative; at the same time, the next evaluation period is extended to 5 days to increase the amount of data accumulated.

[0180] This embodiment extends the generation of tactile guidance signals by introducing a layered progressive guidance strategy, which includes the following steps:

[0181] A tactile guidance intensity grading system is established, including three levels: prompting level, guiding level, and enhancement level. Each level corresponds to a different peak vibration intensity range and timing parameter combination.

[0182] When an intervention-required state is detected for the first time, a prompt-level guidance is initiated. The prompt-level guidance uses a peak vibration intensity lower than the normal guidance intensity, the time interval between adjacent actuators is set to 1.5 to 2 times the optimal time interval, and the duration of a single guidance is short.

[0183] After the prompt-level guidance output, monitor the response of the user's pressure center within the preset observation time window. When the user's pressure center moves in the correct direction but the displacement is less than 50% of the expected target, maintain the prompt-level guidance and extend the guidance duration.

[0184] When the user does not produce a valid response or responds in the wrong direction within the observation time window, the process is upgraded to a guided-level guidance, which uses normal guidance parameters.

[0185] If the user still does not produce a correct response within a preset number of guidance cycles after the guidance output of the guidance level, the guidance is upgraded to the reinforcement level. The reinforcement level guidance increases the peak vibration intensity to 1.2 to 1.5 times the normal intensity, shortens the time interval between adjacent actuators to 0.7 to 0.9 times the optimal time interval, and increases the number of repetitions of the activation sequence.

[0186] During any level of guidance, if it is detected that the user's stress center has moved to the vicinity of the feasible target point, the guidance intensity should be reduced or guidance should be stopped immediately to avoid excessive intervention.

[0187] Record the usage frequency and user response of each level of guidance, dynamically adjust the switching threshold and parameter settings between each level, and optimize the gradualness and adaptability of the guidance strategy;

[0188] If a user remains unresponsive or exhibits significant resistance under enhanced guidance, it is determined that the current moment is not suitable for guidance intervention. Guidance output is paused and the state is recorded. The sensitivity of guidance triggers is reduced in subsequent periods.

[0189] Specifically, the control unit sets up a tactile guidance intensity grading system, including three levels: cue level, guidance level, and enhancement level. Cue level: Peak vibration intensity is 30% to 50% of rated output, and the time interval between adjacent actuators is 1.5 to 2 times the optimal time interval. Guidance level: Peak vibration intensity is 50% to 80% of rated output, and the time interval between adjacent actuators is the optimal time interval. Enhancement level: Peak vibration intensity is 80% to 120% of rated output but does not exceed the safety limit, and the time interval between adjacent actuators is 0.7 to 0.9 times the optimal time interval.

[0190] The peak vibration intensity of the enhanced stage must not exceed 90% of the rated output of the actuator, i.e., 180 millinewtons. This upper limit is determined by the comfort test during the initial calibration phase for the user.

[0191] When the control unit first detects a state requiring intervention, it determines the initial guidance level based on the guidance intensity reference value. Assuming the guidance intensity reference value is 21%, falling within the alert level range, alert level guidance is initiated. The time interval between adjacent actuators at the alert level is set to 150 milliseconds, and the peak vibration intensity is approximately 89 millinewtons.

[0192] After the prompt-level guidance signal is output, the control unit monitors the user's response within the observation window. If the user moves in the correct direction but the displacement is less than 50% of the expected target, the system maintains the prompt-level guidance and extends the guidance duration. If the user does not produce a valid response or the response direction is incorrect, the system escalates to guidance-level guidance. If the user still does not produce a correct response within three consecutive guidance cycles, the system escalates to enhanced guidance.

[0193] During any level of guidance, if the system detects that the user's stress center has moved to the vicinity of the target location, the guidance intensity will be immediately reduced or guidance will be stopped. At the end of each day, the system will analyze the usage of guidance at each level and dynamically adjust the switching thresholds and parameter settings.

[0194] If the user remains unresponsive or exhibits significant resistance under enhanced guidance, the control unit determines that the current moment is not suitable for guidance intervention, suspends guidance output, and records the state. Over the next hour, the control unit reduces the guidance trigger sensitivity, and then gradually restores normal sensitivity after one hour.

[0195] Through the aforementioned layered and progressive guidance strategy, the system can dynamically adjust the guidance intensity based on the user's real-time response, avoiding excessive interference with responsive users while ensuring effective guidance for sluggish users. It also has the ability to identify and respond to user resistance, significantly improving the system's intelligence level and user acceptance.

[0196] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An interactive behavior guidance control method based on execution feedback data, characterized by, include: Collect pressure distribution data from user interfaces and establish a personalized baseline model; Calculate the spatial feature deviation and temporal feature consistency between the current pressure distribution and the personalized baseline model, and output the state classification label and confidence score; Receive state classification labels and confidence scores, assess the constraint violation degree of the current pressure distribution, search for the nearest feasible target point to the user's current state within the feasible domain that satisfies the constraints, and calculate the guiding vector from the current pressure center to the feasible target point; The process of obtaining feasible target points includes: calculating the normalized violation degree of pressure concentration, pressure gradient distribution, pressure ratio between symmetrical regions and effective contact area ratio based on preset ergonomic constraints, and then summing them by weight to obtain the comprehensive constraint violation degree. The stress data is projected into a low-dimensional feature space, and the search is iteratively performed in the opposite direction of the violation gradient until a candidate point that satisfies the constraints and / or has the minimum violation is found. If the low-dimensional Euclidean distance between the candidate point and the current point is less than the user's historical maximum adjustment range, it is confirmed as a feasible target point; otherwise, the step size is adjusted. The system receives the guidance vector, sets the timing interval and intensity envelope according to the pre-calibrated user-personalized parameters, and generates a directional tactile guidance signal; it monitors the direction of movement of the user's pressure center in real time, and adjusts the timing interval and intensity envelope when it detects that the user is moving in the wrong direction and / or has no response for a long time. The intensity of the directional tactile guidance signal is collected as input data and the displacement of the user's pressure center is collected as output data. The second-order system parameters of the user's response characteristics are obtained by recursive least squares method to infer the user's current response state. The process of inferring the user's current response state includes: aligning the tactile guidance signal intensity input and the pressure center displacement output in time to construct a second-order difference equation; minimizing the prediction error using the input-output data sequence, solving the equation coefficients, and extracting damping and gain parameters that characterize the system dynamics; classifying the user's state into a sluggish fatigue state, a responsive alert state, or a resistant state based on the high-low combination of the damping and gain parameters relative to their respective thresholds, and generating adjustment instructions for the control parameters accordingly.

2. The interactive behavioral guidance control method based on execution feedback data according to claim 1, characterized in that, The process of establishing a personalized baseline model includes: extracting the average pressure and standard deviation of fluctuation of each sensing unit from the preset initial time series data, and constructing a spatial pressure distribution baseline map and a spatial fluctuation feature map respectively; identifying the continuous high pressure area in the pressure distribution baseline map as the support area and recording its pressure ratio relationship; extracting the trend reversal time point after low-frequency filtering of the data, and statistically analyzing the reversal interval to obtain the duration range of the user's adjustment action. The spatial pressure distribution baseline map, spatial fluctuation characteristic map, pressure ratio relationship, and adjustment action duration range are stored as the personalized baseline model.

3. The interactive behavioral guidance control method based on execution feedback data according to claim 1, characterized in that, The process of calculating spatial feature deviation and temporal feature consistency includes: based on the spatial pressure distribution benchmark map and the spatial fluctuation feature map, calculating the normalized deviation between the real-time pressure of each sensing unit in the support area and the benchmark value, and statistically analyzing the proportion of units whose normalized deviation exceeds a preset threshold as the spatial feature deviation. Within a preset sliding time window, pressure changes in each support region are assessed to obtain the pressure change trend of each region; the standard deviation of the pressure change rate is calculated to obtain the rate stability index; when the pressure change trends of all support regions are the same and the rate stability index is lower than the preset threshold, the time feature is marked as a consistent gradual change; when the pressure change trends of the support regions are different and / or the rate stability index is higher than the preset threshold, the time feature is marked as an inconsistent change.

4. The interactive behavioral guidance control method based on execution feedback data according to claim 1, wherein, The process of calculating the guidance vector includes: directly extracting the pressure center location information from the low-dimensional feature space coordinates of feasible target points as the target pressure center location; calculating the pressure weighted center coordinates of the current pressure distribution as the current pressure center location; calculating the spatial vector from the current pressure center location to the target pressure center location to obtain the guidance direction; calculating the Euclidean distance between the current pressure center location and the target pressure center location as the deviation distance; mapping the deviation distance to a guidance intensity reference value through a piecewise saturation function; determining whether the guidance direction points to the edge of the effective contact area and / or the recorded unsuitable area; if it points to an unsafe area, correcting the direction to the nearest safe area, and outputting a guidance vector containing the corrected direction and intensity reference value.

5. The interactive behavioral guidance control method based on execution feedback data according to claim 1, wherein, The process of generating the directional tactile guidance signal includes: selecting an activation sequence along the guidance direction in the actuator array; setting the optimal time interval and intensity envelope based on the parameter with the highest user recognition accuracy; generating drive signals for each actuator according to the activation sequence, so that the intensity envelopes of adjacent actuators overlap in time according to a preset ratio; monitoring the pressure center trajectory; increasing the interval and improving the intensity contrast when the angle between the actual movement direction and the guidance direction exceeds the limit; increasing the intensity and shortening the interval when the long-term displacement is insufficient; and / or switching to a single-point continuous vibration mode.

6. An interactive behavior guidance control system based on execution feedback data, which executes an interactive behavior guidance control method based on execution feedback data according to claim 1, characterized by include: The user status recognition module is used to collect pressure distribution data of the user's contact interface and establish a personalized baseline model. Calculate the spatial feature deviation and temporal feature consistency between the current pressure distribution and the personalized baseline model, and output the state classification label and confidence score; The feasible target point calculation module is used to receive state classification labels and confidence scores, evaluate the constraint violation degree of the current pressure distribution, search for the feasible target point closest to the user's current state within the feasible domain that meets the constraints, and calculate the guiding vector from the current pressure center to the feasible target point. The tactile sequence encoding module is used to receive the guidance vector, set the timing interval and intensity envelope according to the pre-calibrated user personalized parameters, and generate a directional tactile guidance signal; monitor the movement direction of the user's pressure center in real time, and adjust the timing interval and intensity envelope when the user is detected to be moving in the wrong direction and / or there is no response for a long time. The parameter co-optimization module is used to collect the intensity of the directional tactile guidance signal as input data and the displacement of the user's pressure center as output data. It obtains the second-order system parameters of the user's response characteristics through the recursive least squares method and infers the user's current response state.

Images